Altis Benchmark Suite

Altis is a benchmark suite to test the performance and other aspects of systems with Graphics Processing Units (GPUs), developed in SCEA lab at University of Texas at Austin. Altis consists of a collection of GPU applications with differnt performance implications. Altis focuses primarily on Compute Unified Device Architecture (CUDA) computing platform.

Documentaion regarding this project can be found at the Wiki page. The Wiki document contains information regarding Altis setup, installation, usage, and other information.

We are refactoring Altis codebase for better usability and making it more developer-friendly. We made sure the benchmark still compile properly during refactoring so you can still use it. The refactoring involves changing how each benchmark application is used and adding more benchmarks.

How to Get Altis

Simply execute the following command:

git clone https://github.com/utcs-scea/altis.git

Prerequisite

Altis relies on the avaialbility of CUDA and CMake (>= 3.8). Please refer to Environment Setup for how to set up Altis.

Build:

After the environment is setup properly, go to the root directory of Altis, simply execute:

./setup.sh

For more information regarding building process, please refer to Build for more information.

To Run a Single Benchmark:

cd build/bin/$BENCHMARK_LEVEL
./$BENCHMARK [options]

General Options: 
    -c, --configFile             specify configuration file
    -d, --device                 specify device to run on
    -i, --inputFile              path of input file
    -o, --outputFile             path of output file
    -m, --metricsFile            path of file to write metrics to
    -n, --passes                 specify number of passes
    -p, --properties             show properties for available platforms and devices (exits afterwards)
    -q, --quiet                  enable minimal output
    -s, --size                   specify problem size
    -v, --verbose                enable verbose output

Note: Run benchmark with –help to see full list of options available for that specific benchmark

Contributing to Altis

Everyone can contribute to Altis, and we value everyone’s contributions. For bugs or proposals, use the issue page. You may also make a pull request directly.

Contact

For any questions regarding this project, please send an email to bodunhu@utexas.edu or rossbach@cs.utexas.edu

Cite Us

If your use of Altos results in a publication, please site:

[1] B. Hu and C. J. Rossbach, “Altis: Modernizing GPGPU Benchmarks,” 2020 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), 2020, pp. 1-11, doi: 10.1109/ISPASS48437.2020.00011.

Bibtex is shown below:

@INPROCEEDINGS{9238617, author={Hu, Bodun and Rossbach, Christopher J.}, booktitle={2020 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)}, title={Altis: Modernizing GPGPU Benchmarks}, year={2020}, volume={}, number={}, pages={1-11}, doi={10.1109/ISPASS48437.2020.00011}}

This work is supported by NSF grants CNS-1618563 and CNS-1846169.